Automatic propagation of manual annotations for multimodal person identification in TV shows | IEEE Conference Publication | IEEE Xplore